Parametric knowledge is information embedded in an LLM’s parameters through training. It includes general facts, language patterns, and reasoning heuristics, and is accessed implicitly during inference. section 5.1
Unlike non-parametric memory, parametric knowledge is static and difficult to update. Agent systems keep user-specific or evolving information outside the model and inject retrieved entries into context when needed. section 1.5.4 section 5.1
Figure 5.1 Memory in an AI agent: parametric memory inside the model, and non-parametric memory kept in persistent and ephemeral storage, surfaced in the context window as long-term memory, short-term memory, and state
Shows parametric memory inside the model alongside external storage and the memory surfaced in an agent’s context window. section 5.1


